Task-oriented dialog systems are computer systems that interact with humans in natural language. The system receives a query, converts the sequence of words into a semantic representation to be used by the dialog manager, decides the best response for the user, and manages the task. Occasionally, the system may receive an out-of-scope query, namely, a query that falls outside the range of the system’s capabilities. In this work, we focus on out-of-scope query prediction, and show how the hierarchical Beta-Bernoulli process outperforms state-of-the-art machine learning classifiers.